Experiments in sensory neurophysiology often record action potential arrival times of nerve cells resulting from spontaneous or stimulus-evoked activity. When all action potentials are taken to be identical and only their localized times of occurrence are considered, one obtains a discrete series of time events characterizing the spike train. It is this series of events that is transmitted down the axon to all of the cell’s targets and that contains most, if not all, of the information that the cell is conveying.
There are two opposing views of neuronal coding, with many intermediate standpoints. One extreme view holds that it is only the average rate, the number of action potentials over some suitable interval usually lasting a fraction of a second or longer, which is relevant for information processing. The opposing view, known as temporal coding , argues that the pattern of spikes, both at the single cell as well as between multiple cells, encodes information. This includes the hypothesis that the exact temporal arrangement of interspike intervals encodes information.
The broad idea of temporal coding is supported by evidence from a variety of sensory systems (locust olfaction, electric fish electrosensation, cat vision and olfaction, monkey vision and audition) pointing towards the role of spike timing, in particular across ensembles of cells, in encoding various aspects of the stimulus.
The characteristics of the neuronal code are closely linked to the seemingly stochastic or random character of neuronal firing. Because little or no information can be encoded into a stream of completely regularly spaced action potentials, this raises the question of how variable neuronal firing really is and of the relation between variability and the neural code. It is the mathematical theory of stochastic point processes and the field of statistical signal processing that offers us the adequate tools for attacking these questions.
Increasingly fast computers and the availability of comprehensive software packages with practical graphical interfaces has made the analysis of neuronal data using such methods more rapid and convenient. One of these packages, Matlab, is well suited for such numerical work. We have implemented a series of data analysis routines for use with spike train data and the corresponding programs (Matlab M-files) can be freely accessed and downloaded from this web-site. We have also implemented several simplified models of spike generation within the simple and intuive programming environment of Matlab, based on the dynamical system simulation package called “Simulink”. Several tutorials introduce the user to the generation of spike train data using these models and to the data analysis. This software is a complement to chapter 9 of the book “Methods in Neuronal Modeling”, edited by Christof Koch and Idan Segev and to be published by MIT Press in 1997. The figures of this chapter which were generated using this package are also described in the following.